摘要
对比序列模式能够表达序列数据集合间的差异,在商品推荐、用户行为分析和电力供应预测等领域有广泛的应用.已有的对比序列模式挖掘算法需要用户设定正例支持度阈值和负例支持度阈值.在不具备足够先验知识的情况下,用户难以设定恰当的支持度阈值,从而可能错失一些对比显著的模式.为此,提出了带间隔约束的top-k对比序列模式挖掘算法k DSP-Miner(top-k distinguishing sequential patterns with gap constraint miner).k DSP-Miner中用户只需设置期望发现的对比最显著的模式个数,从而避免了直接设置对比支持度阈值.相应地,挖掘算法更容易使用,并且结果更易于解释.同时,为了提高算法执行效率,设计了若干剪枝策略和启发策略.进一步设计了k DSP-Miner的多线程版本,以提高其对高维序列元素情况的处理能力.通过在真实世界数据集上的详实实验,验证了算法的有效性和执行效率.
Distinguishing sequential pattern can be used to present the difference between data sets, and thus has wide applications, such as commodity recommendation, user behavior analysis and power supply predication. Previous algorithms on mining distinguishing sequential patterns ask users to set both positive and negative support thresholds. Without sufficient prior knowledge of data sets, it is difficult for users to set the appropriate support thresholds, resulting in missing some significant contrast patterns. To deal with this problem, an algorithm, called kDSP-miner （top-k distinguishing sequential patterns with gap constraint miner）, for mining top-k distinguishing sequential patterns satisfying the gap constraint is proposed. Instead of setting the contrast thresholds directly, a user-friendly parameter, which indicates the expected number of top distinguishing patterns to be discovered, is introduced in kDSP-miner It makes kDSP-miner easy to use, and its mining result more comprehensible. In order to improve the efficiency of kDSP-miner, several pruning strategies and a heuristic strategy are designed. Furthermore, a multi-thread version of kDSP-miner is designed to enhance its applicability in dealing with the sequences with high dimensional set of elements. Experiments on real world data sets demonstrate that the proposed algorithm is effective and efficient.
出处
《软件学报》
EI
CSCD
北大核心
2015年第11期2994-3009,共16页
Journal of Software
基金
国家自然科学基金(61103042)
中国博士后科学基金(2014M552371)
软件工程国家重点实验室开放研究基金(SKLSE2012-09-32)
